Document the ways of using SummaryWriter

[reminisce]

Need to document the recommended way of using SummaryWriter:
with SummaryWriter vs. sw = SummaryWriter(...).

Add an FAQ section in README.

1. How to initialize a SummaryWriter instance: with SummaryWriter(...) or sw = SummaryWriter(..)?
2. What is the meaning of verbose?
3. Should I call flush() on a SummaryWriter object?
4. How to improve the logging performance?
5. Can I use the same tag for adding different types, e.g. image and histogram? No. One of them is going to be overwritten by the other.

[KriteshGarg]

Please add an example how to use summaryWriter() when fit() is used.

[SunDoge]

HI, @reminisce , is there any progress? I'm also wondering about these questions.

[reminisce]

@LordSycho @SunDoge Please refer to https://github.com/awslabs/mxboard/blob/master/demo.md for examples on how to use SummaryWriter.

[SunDoge]

@reminisce I've already known how to use SummaryWriter and I've been using it for a long time. But I'm quite interesting in those five questions. I use default settings all the time and have never thought about them.

Add an FAQ section in README.

1. How to initialize a `SummaryWriter` instance: `with SummaryWriter(...)` or `sw = SummaryWriter(..)`?

2. What is the meaning of `verbose`?

3. Should I call `flush()` on a `SummaryWriter` object?

4. How to improve the logging performance?

5. Can I use the same tag for adding different types, e.g. image and histogram? No. One of them is going to be overwritten by the other.

[reminisce]

@SunDoge Thanks for your interest. I have listed brief answers for 1-4 below, while the 5th has been answered.

1. Using the with statement for SummaryWriter gives you a safe error handling of logging files, which means the logging I/O will be closed properly if some exception is thrown by any code under this statement. It the same philosophy as python's with open(file) statement.

2. verbose=True means printing logging messages. You can turn it off by setting verbose=False.

3. Normally, you don't need to worry about that. When the content is flushed to disk is determined by flush_secs parameter when you initialize SummaryWriter. If you want to see the visualization immediately after data is logged, you will need to call flush().

4. Any I/O operation would possibly result in performance degradation. So you can increase the max_queue and flush_secs to reduce the frequency of writing data to disk to minimize the performance impact of logging, but that would require more memory consumption to hold data for longer time.